Could Intel create an equivalent tech to SLI/Crossfire?

[AusWolf]

The developers still have to do the heavy lifting; no wonder SLI and Crossfire are dead.

It's not worth doing any heavy lifting for maybe 5% of gamers who'd be happy to spend 100% more for a ~50% performance uplift.

Even when SLi and CF worked, buying a faster card was always a better option. I remember wanting to build a Radeon HD 7770 CF system, but it was much cheaper to sell my 7770 and buy a 7970 that not only had 3x more VRAM, but was faster than a 7770 CF system in every single game.

 

[DemonicRyzen666]

It's not worth doing any heavy lifting for maybe 5% of gamers who'd be happy to spend 100% more for a ~50% performance uplift.

Even when SLi and CF worked, buying a faster card was always a better option. I remember wanting to build a Radeon HD 7770 CF system, but it was much cheaper to sell my 7770 and buy a 7970 that not only had 3x more VRAM, but was faster than a 7770 CF system in every single game.

I'm pretty sure it's less than 5% of PC gamers who play only raytracing enabled games, but that hasn't stopped them from being 50% of what DX12 games are.

 

[AnotherReader]

I'm pretty sure it's less than 5% of PC gamers who play only raytracing enabled games, but that hasn't stopped them from being 50% of what DX12 games are.

You're forgetting that two major consoles can also handle limited ray tracing.

 

[ShiBDiB]

Just joining the SLI/CF is dead and never coming back crowd.

It no longer has any practical reason to exist with how powerful even mid-range cards are now.

 

[DemonicRyzen666]

Just joining the SLI/CF is dead and never coming back crowd.

It no longer has any practical reason to exist with how powerful even mid-range cards are now.

Wrong. the RTX 4060 ti needs D.L.S.S & D.L.S.S 3 to even look any better than last gen cards. You know what that it's software support just like what a driver is.

 

[AusWolf]

Wrong. the RTX 4060 ti needs D.L.S.S & D.L.S.S 3 to even look any better than last gen cards. You know what that it's software support just like what a driver is.

Why do you compare to last gen? I compare to my own needs. And for that, a 4060 Ti is plenty. I needed a high-end Radeon HD 7970 only 10 years ago, but now, I'm happy with any last gen midrange card.

 

[Upgrayedd]

SLi stored a mirror image in your VRAM, so with two 4 GB cards, you doubled your theoretical performance (it was closer to 1.5x in practice), but you still only had 4 GB VRAM.

I thought the professional cards with NVLink pooled it?

The developers still have to do the heavy lifting; no wonder SLI and Crossfire are dead.

And with so many moving to UE for the sake of convenience there's not a chance anyone switching would take the time.

Could mGPU be something implemented directly into UE5?

Call me one fashioned but I'd rather have SLI back than DLSS 3 FG.

 

[Shihab]

That whole last state is false.
The options of more choices should matter to everyone especially since it's a feature of DX12 itself. Also because of the obscene prices of newer generation GPU's.

DX12's "feature" is explicit mGPU (which I'm all for), not implicit ones.

And development resources (funding, and most importantly: man hours) are finite. Everyone benefits from stable, efficient drivers and efficient games. But those who can afford the cost and hassle of multi-GPU systems are few.

And you're misinterpreting what I'm saying. I'm not against multi-GPU configs, I'm against the old way of doing it.

 

[masterdeejay]

Vukan can work with multiple GPU-s, explicit mode.
Quake II RTX (steam version)
Intel A770 and A380 detected in console as physical device 0 and 1.

vk_khr_device_group_creation is needed for multi gpu mode
but the driver not enabling it

But it is not working. 0% usage on A380, Both cards support dx12 ultimate and RT.

 

[DemonicRyzen666]

Call me one fashioned but I'd rather have SLI back than DLSS 3 FG.

Gives me an idea to look for a game that actually uses both mGPU & D.L.S.S 3 F.G to compare to my two RTX 2080 T.I's in S.L.I
But what should I be comparing it to in the RTX 4,000 series? the 4060 t.i with D.L.S.S 3 F.G enable?

DX12's "feature" is explicit mGPU (which I'm all for), not implicit ones.

And development resources (funding, and most importantly: man hours) are finite. Everyone benefits from stable, efficient drivers and efficient games. But those who can afford the cost and hassle of multi-GPU systems are few.

And you're misinterpreting what I'm saying. I'm not against multi-GPU configs, I'm against the old way of doing it.

I agree, which is why I hate NVidia for doing the stupid thing of not allowing mGPU without having S.L.I enabled on a few DX12 games. I believe there is only one game to this day that can run two Nvidia cards without having S.L.I enabled for mGPU mode, Ashes of the singularity


But you're missing my point about cost, the point is the cost thing is basically the same it was with dual cards as it with a single card now. The single card RTX 4,000 series is costing around double the price of the same numerical tier they were. It's all because of "software & hardware" One cannot have D.L.S.S 3 F.G without buying into the RTX 4,000 series cards.

I've been against A.F.R which was pushed & praised by reviewers for two things 1. highest possible framerate 2. most even load on both cards. Only now have the decided to drop multi-card support, because of "stutters". Which as we can see today, we still have this problem about a dozen or more newer games still have stuttering problems in them regardless of having a single gpu running them.

Vukan can work with multiple GPU-s, explicit mode.
Quake II RTX (steam version)
Intel A770 and A380 detected in console as physical device 0 and 1.

vk_khr_device_group_creation is needed for multi gpu mode
but the driver not enabling it

But it is not working. 0% usage on A380, Both cards support dx12 ultimate and RT.

View attachment 307493

I might be wrong but seems like it's missing some coding to enable it, or it just might be for two of the same cards & not a imbalanced setup. It might only work with two A770 or two A750 even two A380s.

 

[InVasMani]

Stutter is caused by storage, system memory, cpu cache, and PCIE bus transfer performance reading, copying, and writing data from them over to VRAM. You can alleviate bottlenecks in four ways. Increase capacity of a faster capacity option so you don't run out so easily and fall back on a slower one. Reduce latency so you can access it more responsively. Increase bandwidth so you can transfer more at once. Lastly you can compress the data so you can transmit it more efficiently and/or increase capacity so you don't fall back on a slower option.

The fastest lowest latency one at higher capacity should take priority. Picking between bandwidth, latency, and capacity is mixed depending on usage. How important is capacity or bandwidth or latency in terms of overall system usage and program usage bottlenecks. Focus on the weakest link for more fluid consistency. Focus on the peak performance if you want peak performance. Focus on capacity if you're too near or below typical usage and it leads to slowdown bottlenecks or program crashes from insufficient allocation.

 

[MeohMyoh]

There is real distaste for crossfire and SLI in the general computing public now usually because of a bad experience and a perceived loss of value with the technology. That being said I believe there is a coming use for the technology that will become unavoidable due to the bigger, faster, more featured, larger, faster video card phenomenon will hit a ceiling eventually. Due to Cost wise and of technologically cost to benefit ratios. I myself won't even consider a motherboard that doesn't feature the availability of a x16 single with an X8 by 8x pcie lane assignment capability on a second slot for mgpu usage, And the newer motherboards are fewer and far between nowadays. Even now Amd has gone to chiplet design to enhance their cost to performance ratios. That being said here is a couple of my use experiences with the mgpu combinations

and

also this one